Pip: If you have ever wondered whether someone is quietly engineering what you think by controlling what you say, behavioural science has thoughts — and so does the AI running your search results.
Mara: This episode covers posts from surtymind on the Philosophical Naturalism blog. We are looking at the science of behavioural control, verbal behaviour, and where artificial intelligence fits into that picture.
Pip: Let's start with the question of thought control itself.
Thought Control: Behaviorism, Language, and AI
Mara: The central tension here is whether behavioural science has ever been — or is becoming — a genuine tool for controlling what people say and, through that, what they think.
Pip: And the short answer from early behaviourism was: probably not as much as you feared. The post lays out how rule-following made humans a harder target than a rat on a lever.
Mara: Right. The post puts it directly: "once a creature becomes verbal and starts following rules their behaviour becomes less sensitive to the contingencies of reinforcement." That is the key finding from decades of behavioural research — language is partly a shield against conditioning.
Pip: So Skinner's own toolkit partially inoculates people against Skinnerian control. There is a certain irony there.
Mara: The post traces that arc carefully. Breland and Breland's work on instinctive drift showed non-human animals were less malleable than early behaviourists assumed. Then Sidman and Hayes demonstrated that humans could derive untrained relational frames — coordination, comparison, hierarchy — which made the behaviourally studied human look more like the cognitive scientist's human than the early behaviourist's.
Pip: Which is why public fear shifted. Rogue behaviourists stopped being the nightmare; neuroscientists and AI took their place.
Mara: But the post does not stop at reassurance. It introduces more recent work — Spencer et al. 2022 — identifying techniques to strengthen rule-following compliance: connecting rules to personal values, monitoring pliance rates, and, notably, deemphasizing freedom-threatening language.
Pip: That last one is doing a lot of quiet work.
Mara: It is. And the post connects that to the COVID-19 context, citing Stapleton et al. 2022 on rule-following behaviour during the pandemic as an early real-world application outside therapy.
Mara: The sharpest concern comes at the end: children who grow up perennially online may receive enormous amounts of linguistic input from Large Language Models trained to reflect their designers' values. Not Orwellian word-banning — something subtler.
Pip: Shaping verbal repertoires without anyone noticing the hand on the dial.
Mara: Exactly that. The post's conclusion is that this would not be control of the type Orwell envisioned, but it would be a way of shaping values and verbal habits in ways people may be entirely unaware of.
Pip: So the fear did not go away — it just got a better disguise.
Mara: The underlying question is still Skinner's: if behaviour is always shaped by something, who gets to be the something? That question is only getting harder to answer.
